Apple M2 Ultra and Extreme Design

I am wondering what this means for the high end and desktop performance in general. I can imagine two possibilities:

A. They think what they have is going to be good enough without an “Extreme” version.
B. They don’t care.

I hope for the first, I fear it’s the second.
 
They know their roadmap for the SoC designs, and probably by the time we get to M3 Ultra it will be way faster than M1 Ultra - at some point the need for more and more cores drops off. Also they may be thinking that there is insufficient demand for something like the Extreme to justify the fab resources.
 
There are still these options to distinguish their desktops:

1) Higher single-core CPU performance through increased clocks.

2) Higher GPU performance through separate GPU-only SoC's (as others have mentioned, if they do this, it probably won't be for the MP only, but will be a general approach used on all their desktops, and is thus something we won't see for a while, if at all).

3) And for the MacPro specifically: All the computing components for the 2019 MP are on a single board that can be easily slid in and out. I'm assuming it's the same for the AS MP. If so, why not make it generationally upgradeable? It ticks lots of boxes: (a) it's green; (b) it's modular; (c) it rewards customers who've invested the extra money for the MP case; (d) it just makes sense--most of the case components should last a decade plus, and the fans and power supply should be overbuilt for all future generations of AS (since they were designed to handle the power requirements of Xeon/Radeon).
 
Last edited:
1. From somebody who works on Apple's GPU team.
2. The quad chip has been canned, which our resident CPU engineer corroborates.
3. Right now, they are working on what will be the M5 chip.
4. Quad chip was only ever specced for the M1 series and removed late in the project.
5. No plans to resurrect the quad design through the M7 generation.
6. Quad was too much effort for too small a market.
7. Multi-die packaging may come with the M8 or later generations, which allows for CPU and GPU to be fabbed on separate dies. However, no such plans currently exist.

Look man, I just want a Mac Cube with a Mn Extreme... ;^p
 
Look man, I just want a Mac Cube with a Mn Extreme... ;^p
Hey, all you have to do is stack a couple of Studio Ultras to approximate a cube shape and set up networked process sharing between them and voilá, a Cube Extreme!
 
It's kind of interesting since a depiction of quad configuration is featured in a recent Apple patent:


Of course, it doesn't need to mean anything, and it could be just an artefact from the early Extreme era, but I do find it notable that all the previous patents only ever depicted two SoCs (with a direct connection) while this one depicts four SoCs with a routing network between them.

It does make sense that Extreme is too much effort though. I wonder whether they have any other plans for scaling to the needs of high-end desktop or whether they are just going to drop it.

I also wonder where this leaves this type of multi-die product, especially since @Colstan writes that no multi-die packaging is planned any time soon:


I thought it would be a great way to deal with the increased cost of 3nm process while increasing both the compute density and caches in a SoC. I am a bit worried about Apple's ability to continue pushing the performance boundaries if they stay with the monolithic die.
While I didn’t expect the “let’s build legos” packaging to come in M3, I’m surprised that this rumor, if true, states that the current packaging scheme will be extant for what, ten+ years? M1 through basically M7? Maybe even later? I was under the impression things were moving faster than that in packaging tech.
 
While I didn’t expect the “let’s build legos” packaging to come in M3, I’m surprised that this rumor, if true, states that the current packaging scheme will be extant for what, ten+ years? M1 through basically M7? Maybe even later? I was under the impression things were moving faster than that in packaging tech.
I wouldn't have paid this rumor any mind if it hadn't coincided with @Cmaier noting the quad getting canned at the same moment, and he does find what this person says to be believeable. From the wording of the source that Snell and Hurley are quoting, Apple is working on or have planned everything through the M7 generation. That source states that multi-die packaging could come with M8 or later, but that's speculation on the source's part, simply because Apple haven't mapped out the M8+ roadmap. The fruit company must see value in staying with a fully integrated die and not going chiplet like their competitors. Also, unlike incompetent Intel and perennial underdog AMD, they've got the funds to not worry about needing to go hodgepodge if they don't want to.

I find this rumor to be entirely believable, the "Extreme" chip was always wish casting that Apple knew wasn't worth the effort, and by the time the M3 generation has been fully released, we'll be able to establish a proper cadence and product portfolio for Apple Silicon. I'm sure most folks here noted that I've gotten tired of the speculation that amounted to nothing more than nerd fever dreams. I'm not blameless, at one point I took part in the nonsense fiesta myself, but it'll be good to have multiple generations of actual products to base informed speculation upon.

Sure, I'd like to see Apple resurrect the quad to go inside a $12,000 Mac Pro that nobody will buy. I'd also like to have a purple unicorn that craps gold, but I'm not going to get one of those, either.
 
I wouldn't have paid this rumor any mind if it hadn't coincided with @Cmaier noting the quad getting canned at the same moment, and he does find what this person says to be believeable. From the wording of the source that Snell and Hurley are quoting, Apple is working on or have planned everything through the M7 generation. That source states that multi-die packaging could come with M8 or later, but that's speculation on the source's part, simply because Apple haven't mapped out the M8+ roadmap. The fruit company must see value in staying with a fully integrated die and not going chiplet like their competitors. Also, unlike incompetent Intel and perennial underdog AMD, they've got the funds to not worry about needing to go hodgepodge if they don't want to.

I find this rumor to be entirely believable, the "Extreme" chip was always wish casting that Apple knew wasn't worth the effort, and by the time the M3 generation has been fully released, we'll be able to establish a proper cadence and product portfolio for Apple Silicon. I'm sure most folks here noted that I've gotten tired of the speculation that amounted to nothing more than nerd fever dreams. I'm not blameless, at one point I took part in the nonsense fiesta myself, but it'll be good to have multiple generations of actual products to base informed speculation upon.

Sure, I'd like to see Apple resurrect the quad to go inside a $12,000 Mac Pro that nobody will buy. I'd also like to have a purple unicorn that craps gold, but I'm not going to get one of those, either.
The Extreme wasn’t really a nerd fever dream as even according to the reports that it’s dead Apple themselves reportedly planned on it and to be frank the Extreme is the only reason for the Pro Tower to exist. What was released is the equivalent of the 13” pro - a product released out of rote habit, out of an expectation that it should exist, but not one they believe in anymore.

Further an Extreme Mac Pro need not have cost $12k. While Apple’s upgrade pricing leaves something to be desired, until the release of the Mac Pro, their baseline pricing of the AS lineup was shockingly good to decent. The Mac Pro meanwhile is almost priced to down sell customers to the Studio - a “seriously don’t buy this, no really get the Studio” price.

Now maybe they’re right, workstations may be a dead product segment, not worth pursuing anymore. I’m a little surprised though because a) halo products move lower stack products and b) despite being low volume they tend be high profit. But maybe not anymore, too much of that kind of work is simply done in the cloud/cluster which Apple abandoned/never really participated in. The future of that market segment is shrinking too much to be viable - pressed on the lower end by the high end consumer machines and on the high end by the aforementioned cluster/cloud. That situation leaves the more traditional workstation tower in a precarious position. But then Apple released a Pro anyway, despite their own belief according to this rumor that the product has no real future and cannot justify its existence against the lower end of the product stack.

Apple’s pockets are deep, but they aren’t infinite as evidenced by the fact that even they produce chiplets for their products, eg the Ultra. And to be frank, they run very lean on engineering staff, which is another possible cause of not being able to produce an Extreme. There are multiple reports of the SOC team after the M1 being happy with their products but thoroughly exhausted getting out multiple new SOC lines with a lack of new people requiring a massive input from existing staff. Another SOC even one originally planned to be 4 smaller ones done together may have simply been beyond the team’s resources to pull off.

Further, the benefits of the chiplet approach extend beyond cost savings. And while it’s also thoroughly possible that the packaging technology may simply not be ready yet, Apple’s own patents, strings found in Apple device codes, and previous rumors indicated a switch to greater use. This was not just pie in the sky thinking either.
 
Last edited:
I wouldn't have paid this rumor any mind if it hadn't coincided with @Cmaier noting the quad getting canned at the same moment, and he does find what this person says to be believeable. From the wording of the source that Snell and Hurley are quoting, Apple is working on or have planned everything through the M7 generation. That source states that multi-die packaging could come with M8 or later, but that's speculation on the source's part, simply because Apple haven't mapped out the M8+ roadmap. The fruit company must see value in staying with a fully integrated die and not going chiplet like their competitors. Also, unlike incompetent Intel and perennial underdog AMD, they've got the funds to not worry about needing to go hodgepodge if they don't want to.

I find this rumor to be entirely believable, the "Extreme" chip was always wish casting that Apple knew wasn't worth the effort, and by the time the M3 generation has been fully released, we'll be able to establish a proper cadence and product portfolio for Apple Silicon. I'm sure most folks here noted that I've gotten tired of the speculation that amounted to nothing more than nerd fever dreams. I'm not blameless, at one point I took part in the nonsense fiesta myself, but it'll be good to have multiple generations of actual products to base informed speculation upon.

Sure, I'd like to see Apple resurrect the quad to go inside a $12,000 Mac Pro that nobody will buy. I'd also like to have a purple unicorn that craps gold, but I'm not going to get one of those, either.

I wouldn’t put much stock on any planning that is for a product more than 2 years out. Nobody is working on M6 or M7 and, from experience, internal roadmaps are meaningless other than as exercises in forecasting needed man-power and EDA licenses. Once you start working on a chip, at that point the system architecture is locked down. But M5 may be very different than what this source assumes it is, and M7 is not even worth speculating on.
 
The Extreme wasn’t really a nerd fever dream as even according to the reports that it’s dead Apple themselves reportedly planned on it and to be frank the Extreme is the only reason for the Pro Tower to exist.
I apologize, I should have specified that I wasn't talking about the "Extreme". In fact, I had been defending Gurman months ago when he said that it was planned for the M1 generation. It's not until recently that he lost the plot and burned his best sources. The nerd fever dreams were things like compute modules, GPGPU, third-party graphics cards, slotted RAM, swappable CPUs, all consumed with a heavy dose of hopeium.

I don't claim to be some Svengali with my magic crystal ball. I simply predicted the most boring, staid product that Apple could release as a Mac Pro, and stumbled, staggered, and fell on top of the correct answer. I'm thoroughly enjoying my first place prize of cold pizza and bitterness.

Further an Extreme Mac Pro need not have cost $12k.
That was hyperbole, my friend. It doesn't matter what it might cost because it doesn't exist.

The Mac Pro meanwhile is almost priced to down sell customers to the Studio - a “seriously don’t buy this, no really get the Studio” price.
I think the Mac Pro is suffering from the same fate as the iMac, just worse. The iMac was once the crown jewel of the consumer Mac line. Now, Apple can't be arsed enough to update it with an M2. The Mac Pro was the fabulous bejeweled halo product, now it's the expensive gewgaw that Apple hides in the middle of a presentation, perhaps only updated because Ternus said it was "for another day". They already have the fancy case and Ultra chip, might as well slap it inside and call it a day.

Phil Schiller once said that every product has to justify its existence. Here's how Phil explained the Mac Pro back in 2017:

Mac Pro is actually a small percentage of our CPUs — just a single-digit percent. However, we don’t look at it that way. The way we look at it is that there is an ecosystem here that is related. So there might be a single-digit percentage of pros who use a Mac Pro; there’s that 15 percent base that uses Pro software frequently and 30 percent who use it casually, and these are related. These are not distinct little silos. There’s a connection between all of this.

Courtesy of Jason Snell at Macworld, back in February of this year. I think Snell put it better than I could:

That’s Schiller explaining that the Mac Pro is valuable because… well, because it’s connected to the people who use Pro software a little and who use Pro software a lot, and… it’s all related, I guess? It sure seems a lot squishier when you think about it.

Phil's word salad from six years ago pretty much shows how difficult it has been for Apple to justify the Mac Pro's existence, which I think is why the quad "Extreme" got axed during M1 development.

But then Apple released a Pro anyway, despite their own belief according to this rumor that the product has no real future and cannot justify its existence against the lower end of the product stack.

I think Apple was still kicking around the idea of perhaps having another go at the quad, back when Ternus made his comment. After Ternus' vague statement, Apple had no choice but to make an Apple Silicon Mac Pro, but it could be the last one. It depends on how many users still need slots for their non-GPU tasks. It's no longer a hotrod race car, but a truck sporting pink flamingo mud flaps.

Apple’s pockets are deep, but they aren’t infinite as evidenced by the fact that even they produce chiplets for their products, eg the Ultra.
Apple's pockets are deep because they don't waste dosh on frivolous vanity projects. As Steve Jobs once said, "real artists ship". They could have engineered the quad, it just didn't make good business sense, which includes not using valuable engineering resources on a handful of niche customers.

And to be frank, they run very lean on engineering staff, which is another possible cause of not being able to produce an Extreme. There are multiple reports of the SOC team after the M1 being happy with their products but thoroughly exhausted getting out multiple new SOC lines with a lack of new people requiring a massive input from existing staff.
That's absolutely true. We've often heard reports of how relatively small Apple's engineering teams are. Quality CPU architects don't grow on trees. Sometimes those invaluable engineers move to other companies, retire, get hit by a bus, or become patent attorneys.

Further, the benefits of the chiplet approach extend beyond cost savings. And while it’s also thoroughly possible that the packaging technology may simply not be ready yet, Apple’s own patents, strings found in Apple device codes, and previous rumors indicated a switch to greater use. This was not just pie in the sky thinking either.
I don't think chiplets are a panacea, I don't place it in the circus side show category. I just don't think they are necessarily the next best thing, particularly when Apple has access to the best nodes and can integrate with impunity. I'm ambivalent about whether they go the multi-die route or not. I trust Johny Srouji to make the right call on that, I just want good products.

The patents may be interesting, but patents ain't products. I'm content with waiting until Apple actually ship a physical product, and leave the arcane wizardry of patent black magic to folks who are much smarter than I am.

That being said, don't tell me this doesn't look like it was made in a clown car:

intelpat.jpg


Regardless, I don't enjoy being a buzzkill. I'm simply not one for wild speculation. I appreciate your eternal optimism, @dada_dave. I don't know which garden they grow such people as you, but I'm glad that they do. Keep in mind that, here at TechBoards, the week isn't officially over until Colstan has posted his weekly rant about whatever randomly comes to mind.

I wouldn’t put much stock on any planning that is for a product more than 2 years out. Nobody is working on M6 or M7 and, from experience, internal roadmaps are meaningless other than as exercises in forecasting needed man-power and EDA licenses. Once you start working on a chip, at that point the system architecture is locked down. But M5 may be very different than what this source assumes it is, and M7 is not even worth speculating on.
Noted. As always, I appreciate, respect, and defer to your expertise.
 
I am a bit worried about Apple's ability to continue pushing the performance boundaries if they stay with the monolithic die.
Are you thinking about the size of the Max chip, and thus the real estate available for more GPU cores and larger coprocessors, being limited by the reticle?
 
Are you thinking about the size of the Max chip, and thus the real estate available for more GPU cores and larger coprocessors, being limited by the reticle?

I was thinking more about rising costs for new processes and less than optimal
improvements in SRAM scaling. Making chips at the reticle limit is prohibitively expensive, not to mention that low yields will mean limited ability to supply enough devices.

But again, it’s all just empty speculation. I have no industry knowledge or experience. I would just really want to know what Apples long term plans are in regards to performance-oriented products.
 
But again, it’s all just empty speculation. I have no industry knowledge or experience.
I have nothing of value to add to this highly technical discussion. I just wanted to say that I thoroughly enjoy whenever @leman says "I only have amateur knowledge", and then proceeds to prove that the rest of us are the real amateurs here. I appreciate your humility, but you constantly downplay your big brain.
 
The stuff about the Extreme dying makes me wonder a little if Apple has demographic research data and it shows them that most (large majority) of the Mac Pro audience are now both using and happy with the Studio rocking the Ultra? There is a market size below which the fab runs are too small to be economically feasible.
 
I could, maybe, see a 3-piece chip composed of the CPU+ANE part, the special function logic (like comm units, the H.264 accelerator and so forth) and the GPU, connected with ultrafusion so that chips could be built as needed. It would be more expensive than a basic SoC from a single burn, and perhaps impractical, but it might be better for getting good yields and might possibly get more stuff onto a single wafer.
 
The reviews of the Mac Pro are as expected - bottom line: unnecessary and unnecessarily expensive. :( Reviewers could only find a handful of cards (and that’s being generous) where it made sense to use the actual internal PCIe connection as opposed to putting the Studio into a chassis or using an external enclosure for a lot less money with otherwise no improvement in performance, heat, or noise.
 
The reviews of the Mac Pro are as expected - bottom line: unnecessary and unnecessarily expensive. :( Reviewers could only find a handful of cards (and that’s being generous) where it made sense to use the actual internal PCIe connection as opposed to putting the Studio into a chassis or using an external enclosure for a lot less money with otherwise no improvement in performance, heat, or noise.
LTT review?
 
Back
Top